Impact of RAN Virtualization on Fronthaul Latency Budget ...F. Giannone1, H. Gupta2, K. Kondepu1, D....
Transcript of Impact of RAN Virtualization on Fronthaul Latency Budget ...F. Giannone1, H. Gupta2, K. Kondepu1, D....
F. Giannone1, H. Gupta2, K. Kondepu1, D. Manicone1, A. Franklin2, P. Castoldi1, L. Valcarenghi1
1Scuola Superiore Sant’Anna, Pisa, Italy2Indian Institute of Technology Hyderabad, India
Impact of RAN Virtualization on Fronthaul Latency Budget: An Experimental
Evaluation
WS-21: International Workshop on 5G Test-Beds & Trials – Learnings from implementing 5G (5G-Testbed 2017)
Globecom 2017, 4-8 December 2017 – Singapore, Singapore
© 2017 Scuola Superiore Sant’Anna
2
Summary
• Virtualized New Radio Access Network (RAN)• Virtual Network Function (VNF) performance evaluation• Implementation of EPC and RAN functions in ARNO-5G
testbed• Experimental Results• Conclusions
© 2017 Scuola Superiore Sant’Anna
3
Virtualized New RAN
• RAN split– Fronthaul– Backhaul
• gNB functional split– Distributed Unit (DU)
– Central Unit (CU)
Fronthaul
Backhaul
Micro-cloud/fog node
Cloud node
gNB
© 2017 Scuola Superiore Sant’Anna
4
gNB Functional Splits (3GPP TR 38.801)
Option 8Option 7-1
© 2017 Scuola Superiore Sant’Anna
5
Fronthaul requirements (TR 38.801)
Not yet clarified
due to 4msHARQ process
Source: LTE for UMTS: Evolution to LTE-Advanced, 2nd Edition, Harri Holma, Antti Toskala ISBN: 978-0-470-66000-3
© 2017 Scuola Superiore Sant’Anna
6
Study objective
1. Are the extra levels of abstraction impacting the fronthaul latency constraints?
2. Does the jitter impact the fronthaul link performance?
© 2017 Scuola Superiore Sant’Anna
7
Functional element placement in the ARNO-5G testbedarnotestbed.santannapisa.it
Intel i7 4790 (@ 3.60 GHz)
PC1
CiscoCatalyst 2960G
PC2
PC7
PC4 PC5
PC6
LTEHuawei E3372
© 2017 Scuola Superiore Sant’Anna
8
Mobile Network Software – OpenAirInterface
Core Implementation
• openair-cn• Implements the EPC 3GPP specs• Contains the implemention of:
Home Subscriver Server (HSS) Mobile Management Entity (MME) Serving Gateway (S-GW) PDN Gateway (PDN-GW)
RAN Implementation
• openairinterface5g
• Implementation of Rel 10 LTE of: Evolved NodeB (eNB); User Equipment (UE).
• Implemented functional splits options: IF4p5 Option 7-1 (intra-PHY split) IF5 Option 8 (PHY-RF split)
Option 7-1
Uplink Direction Downlink Direction
DU FFT, CP removal and PRACH filtering
IFFT, CP addition and PRACH filtering
CU Rest of PHY functions and the higher layers
Rest of PHY functions and the higher layers
• For the Core and RAN implementation the OpenAir interface software is used.
© 2017 Scuola Superiore Sant’Anna
9
Performance Evaluation Parameters• Allowable latency budget: allowable one-way fronthaul latency between DU and
CU
• Allowable jitter budget: allowable one-way fronthaul jitter between DU and CU
© 2017 Scuola Superiore Sant’Anna
10
Emulations of Latency and Jitter on the fronthaul• The linux utility traffic control ¨tc-netem¨ is used
• A delay d0 is applied to the DU Ethernet interface towards the CU.
• A delay d1 is applied to the CU Ethernet interface towards the DU.
• Evaluation of the frontahul latency budget: d0 and d1 are increased with
steps of 10 µs until DU, CU and UE disconect.
• Evalution of the fronthaul jitter budget: A jitter following a normal
distribution is added to the latency values d0 and d1 with standard deviation increased of steps of 5µs.
Source: https://www.excentis.com/blog/use-linux-traffic-control-impairment-node-test-environment-part-2
© 2017 Scuola Superiore Sant’Anna
11
Considered scenarios• Scenario 1:
• All functional elements deployed in bare metal
• Scenario 3:• CU and EPC virtualized
through VirtualBox• DU in bare metal• All functional elements
based on juju charms and managed through Juju.
• Scenario 2:• Two DUs in bare
metal connected to two instances of CU running in the same bare metal
© 2017 Scuola Superiore Sant’Anna
12
Allowable Latency Budget
• Allowable latency budget always below 250 μs• Allowable latency budget decreases if the signal bandwidth and if the number of DUs
connected to the same CU increases due to heavier processing• Allowable latency budget is much lower if mobile network functions are virtualized
© 2017 Scuola Superiore Sant’Anna
13
Impact of Allowable Jitter Budget on Allowable Latency Budget
• Jitter negatively impacts latency budget
220 μs
160 μs 200 μs
120 μs
30 μs
• The Jitter is applied to a latency value close to the fronthaul allowable latency budget• The fixed latency value is choosen according to the scenario
© 2017 Scuola Superiore Sant’Anna
14
Allowable Jitter Budget
• Jitter negatively impacts fronthaul because there are periods in which not enough samples (i.e., modulation symbols) can be delivered to the PHY layer
100 μs
50 μs 20 μs
100 μs
50 μs
• The Jitter is applied to a latency value far to the fronthaul allowable latency budget• The fixed latency value is choosen according to the scenario
© 2017 Scuola Superiore Sant’Anna
15
Conclusions
• Experimental evaluation of the impact of virtualizing eNB functions on the fronthaul latency and jitter budget
• Functional split Option 7-1 (i.e. intra-PHY) and Option 8 (PHY-RF) are applied
• No virtualization• Virtualization based on Virtualbox• The fronthaul latency bandwidth reduction depends on the considered signal
bandwidth (i.e. 5 MHz, 10 MHz) and on the number of functions running in the same device
• Virtualization decreases the allowable latency budget• A jitter of at most 40 us can be tolerated
© 2017 Scuola Superiore Sant’Anna
16
email: email: [email protected] [email protected]
thank you!thank you!
This work has been partially funded by the EU H2020 5G-Transformer Project (grant no. 761536)
© 2017 Scuola Superiore Sant’Anna
17
Backup SlidesBackup Slides
© 2017 Scuola Superiore Sant’Anna
18
gNB Virtualization• ETSI (ETSI GS NFV-PER 001
V1.1.1 (2014-06)) descriptors:– Virtual Network Function (VNF)
– Virtual Machine (VM)
– Compute Host
• What virtualization implies: applications running in the guest
host have “to cross” several layers of abstraction.
Extra levels of abstraction reduce workload performance.
• Different virtualization types: Hypervisor-based virtualizations:
allow to fully emulate a CPU architecture and OS;
Container-based virtualizations: utilizes kernel features to create an
isolated environment of the process using the host hardware.
© 2017 Scuola Superiore Sant’Anna
19
Fronthaul requirements (TR 38.801)
Not yet clarified
ASSUMPTIONS
due to 4msHARQ process
To be reviewed
Source: LTE for UMTS: Evolution to LTE-Advanced, 2nd EditionHarri Holma, Antti Toskala ISBN: 978-0-470-66000-3
© 2017 Scuola Superiore Sant’Anna
20
Virtualized EPC and CU Network configuration
• SPGW-VM: 1 Virtual Interfaces in host-only networking mode (S11 interface); 1 Virtual Interfaces in bridge networking mode (S1-U interface)
• CU-VM: 1 Virtual Interfaces in bridge networking mode (S1-U and S1-C interfaces); 1 Virtual Interfaces in bridge networking mode (fronthaul link).
• HSS-VM: 1 Virtual Interfaces in host-only networking (s6a interface);
• MME-VM: 1 Virtual Interfaces in host-only networking mode (S6a interface); 1 Virtual Interfaces in bridge networking mode (S1-C interface);